One line that halves them both

Two shapes are lying on a table. They are not the same size, they are not the same shape, and nobody arranged them. The claim is that a single straight cut divides each of them exactly in half.

Nothing about the two shapes was chosen to make this work, and nothing about them could be. The result holds for any two regions with a finite area — two puddles, two countries, a slice of bread and a slice of ham — and it holds for the same reason in every case, which is a reason with no geometry in it at all.

A quantity that reverses sign #

Start one dimension down, where the whole argument is visible in a single picture.

Take any continuous quantity defined around a circle: a temperature, a height, a shade of grey. Read it at a point, then read it at the point directly opposite. The two readings will usually differ. The claim is that some pair of opposite points gives the same reading twice.

The proof is one line long. Let g be the difference between the reading at a point and the reading at its opposite point. Then g at the opposite point is minus g at the original point, because the same two numbers are being subtracted the other way round. So g takes a value and also takes its negative. A continuous quantity that is positive somewhere and negative somewhere else is zero somewhere between, and where g is zero the two opposite readings agree.

That last step is the intermediate value property, and it is doing all of the work. It is worth being precise about what it needs: the quantity must be defined everywhere on the circle and must not jump. Continuity is the whole of the hypothesis, and a quantity that jumps has no obligation to obey.

The figure finds the crossing by bisection rather than pointing at it — halving the interval two hundred times — and then checks that the two readings really are equal to twelve decimal places. That distinction matters throughout this argument, and it will come back at the end: knowing a thing exists and knowing where it is are separate pieces of knowledge, and this proof supplies only the first.

From one circle to two shapes #

Now the plane. Fix a direction, and consider all the lines perpendicular to it. Sliding such a line across the first shape sweeps its area smoothly from nothing to all of it, so at exactly one position the line has half the shape on either side. Every direction therefore has one line attached to it: the line, perpendicular to that direction, that halves the first shape.

That line does something to the second shape as well. It leaves some fraction of it on the near side and the rest on the far side, and the difference between those two amounts is a number that depends on the direction chosen.

Turn the direction through a half turn. The perpendicular line is the same line — a line has no direction of its own — so it still halves the first shape. What changes is which side is called near and which far, so the imbalance in the second shape becomes exactly its own negative.

That is the same g as before, defined on the circle of directions rather than on a circle drawn on a page. It reverses sign between opposite points, so it is zero somewhere, and where it is zero the line halves the second shape as well as the first.

The whole argument is that one sentence, applied twice.

The search, and what it is allowed to conclude #

The sweep in the figure above is not the proof. It is a search, and it is worth separating the two, because the site’s habit is to make a picture decide something and this is a case where the picture decides less than the argument does.

What the sweep does: it computes the imbalance at three hundred and sixty directions, notices a sign change between two of them, and then bisects sixty times to land on the crossing. What it produces is a direction good to about a millionth of a degree, at which both areas are halved to nine decimal places.

What the sweep cannot do is establish that a crossing exists. A finite scan of a continuous quantity can always miss a feature between its samples — a curve that dips below zero and returns between two adjacent readings leaves no trace at all in the samples. The existence comes from the sign reversal, which is exact and needs no sampling: the imbalance at one direction and at its opposite are the same number with opposite signs, and that is arithmetic rather than measurement.

So the figure’s honest claim is the narrow one: here is a cut, and here are its four areas. The theorem’s claim is the wide one: a cut exists, whatever the shapes. The first is checkable and the second is not, which is why the second needs a proof.

Why an odd number cannot be zero #

Underneath the sign-reversal argument is a more structural one, and it is the version that generalises.

A map that sends every point of a circle to a point of a circle can be asked how many times it goes round — the same winding count that decides how many roots a polynomial has. Now suppose the map sends opposite points to opposite points. Then it must go round an odd number of times.

The reason is easy to feel and slightly fiddly to write: going halfway round the domain moves the image half a turn, or three half-turns, or five, but always an odd number of them, so going all the way round accumulates twice an odd number of half turns, which is an odd number of whole turns.

An odd number is never zero. And a map that could be shrunk continuously to a constant would have to have count zero, because a constant map goes round no times at all and the count cannot change under a deformation. So a map sending opposite points to opposite points can never be shrunk to a point — which is exactly the statement that no continuous map of the disc to its boundary circle can be antipodal on that circle, and one rearrangement later, the statement that the difference g must vanish.

That argument is Brouwer’s in disguise, and the two theorems are usually proved together for that reason. The invariant that cannot change is the whole mechanism in both.

The relationship between the two runs one way cleanly and the other way with effort. Given Borsuk–Ulam, Brouwer follows in a few lines; given Brouwer, the antipodal statement is harder to reach, because Brouwer says nothing about opposite points and the extra symmetry has to come from somewhere. Both are consequences of the same fact about winding counts, and both are usually presented as the first serious dividend of that fact.

Two conditions, and one that is not needed #

The theorem asks for very little, and it is worth naming exactly what.

The regions need an area that behaves. A shape whose area cannot be sliced continuously — a set so pathological that the mass on one side of a moving line jumps — falls outside the argument. Any region a reader can picture is fine; the caution is for the sets a measure theorist can build and nobody can draw.

The regions must be bounded. An unbounded region can have half of an infinite area on both sides of every line, in which case “half” means nothing.

What is emphatically not needed is any relation between the two shapes. They may overlap, one may lie inside the other, they may be in pieces. A shape in three separate pieces is halved by the same argument, because nothing in it referred to the shape being connected — a fact that is easy to miss and immediately doubles the theorem’s reach.

Nor do they need to be convex. The figure uses two convex blobs because a convex shape’s halves are easy to shade, not because convexity is used anywhere. For a non-convex shape the halving line still exists and the two halves it produces may each be in several pieces, which changes nothing: the area on one side of a line is a well-defined number whether or not the region on that side is connected.

One more thing is not needed, and its absence is the reason the statement is not obvious. Nothing requires the two shapes to be anywhere near each other. Two blobs a mile apart are halved by one line just as surely as two overlapping ones, and the line in that case is nearly parallel to the segment joining them — which a reader who has been imagining shapes side by side may find worth drawing on the back of an envelope before believing.

Three dimensions, and the sandwich the theorem is named for #

The plane version halves two regions with a line. The space version halves three regions with a plane, and this is the statement the result is usually quoted as: given a slice of ham, a slice of bread and a second slice of bread, however placed, one straight cut through all three leaves half of each on either side of the knife.

The argument is the same one, run on the sphere of directions instead of the circle. Each direction gives the plane perpendicular to it that halves the first solid; that plane leaves the other two out of balance by two numbers; and those two numbers, taken together as a point of the plane, reverse under swapping to the opposite direction. The Borsuk–Ulam theorem in its general form says a continuous map from the n-sphere to n-space sends some pair of opposite points to the same value, and applying it here gives a direction at which both imbalances vanish at once.

The two-dimensional case of that statement has a striking reading of its own: at every moment there are two points directly opposite each other on the Earth’s surface with the same temperature and the same pressure. Two quantities, one sphere, one pair of points — and no measurement anywhere in the derivation.

What it costs #

Nothing in the proof says where the cut is, and the computation is genuinely harder than the proof.

For two convex polygons the cut can be found by the sweep above, and the cost is a bisection on the direction with a bisection on the offset inside it — a few hundred area computations for a millionth of a degree. For point sets rather than regions the problem changes character: the ham-sandwich cut of two finite sets of points is a line with half of each set on each side, and finding it exactly is a combinatorial problem on the arrangement of lines through pairs of points, which is where the algorithmic literature lives.

The gap between the two costs is the usual one for this kind of theorem. A proof by sign change is short because it never has to construct anything, and every construction afterwards has to be paid for separately. Nonconstructive existence is cheap and specific answers are not.

The same trick, two fields away #

The imbalance argument turns up wherever a decision has to be made about two competing quantities at once, and the closest neighbour on this site is not in topology at all.

Cut-and-choose works because the cutter’s own valuation of the left piece minus the right piece is a continuous quantity that runs from positive to negative as the knife slides, so somewhere it is zero. The moving-knife procedures for three people are the same argument again, with more knives. What the topology adds is the ability to run the argument in several quantities at once, which is precisely what “halve two things with one cut” needs.

The connection is not decorative. The necklace-splitting theorem — two thieves dividing a necklace of several kinds of bead with as few cuts as possible — is proved from Borsuk–Ulam directly, and the number of cuts it needs is the number of kinds of bead. Fair division and antipodal points are the same subject wearing two hats.

What the picture cannot show #

The figures on this page draw two shapes and one line, and the theorem is about all shapes and no particular line. That gap is not a defect in the drawing; it is the difference between an instance and a statement, and no single figure closes it.

There is also something the sweep hides. The curve of imbalance in the second figure crosses zero at least twice — it must, being its own negative half a turn later — so there are always at least two halving cuts, and generally an even number of them. The figure marks one. Which one it marks depends on where the scan happened to find its first sign change, and nothing distinguishes the cuts from each other.

And the drawing shows areas, which is the easiest case. The theorem is really about measures, and a measure need not be an area: it can be a population, a mass, a probability. The picture cannot show a cut halving two populations, because a population has no shape, and the argument does not care.

The ladder from here #

Rungs above: the Borsuk–Ulam theorem stated and proved on the sphere, with the antipodal map’s degree doing the work. The Lusternik–Schnirelmann covering theorem, which is the same result in the form “cover a sphere with three closed sets and one of them contains a pair of opposite points”. The ham-sandwich theorem in n dimensions, and the way its proof reduces to the sphere. Necklace splitting, and the count of cuts. The discrete version for finite point sets, with its own algorithm and its own cost. Tucker’s lemma, which is the combinatorial shadow of the whole thing and is proved by counting rather than by continuity. And the connection to Brouwer run the other way — Borsuk–Ulam implies Brouwer, and the derivation is three lines.

Why the argument is worth keeping #

A great many existence proofs on this site have the same shape: a quantity is shown to take two signs, and continuity supplies the zero. It settles that a polynomial has a root, that a map of the disc fixes a point, that something on a sphere fails to be combed, that a fair cut exists.

What separates this one is how little it assumes about what is being halved. There is no requirement that the shapes be nice, no requirement that they be related, no requirement that either of them be connected. Two arbitrary bounded regions, one straight line. The theorem’s reach comes from the poverty of its hypothesis, and that is a pattern worth recognising: the results that apply everywhere are the ones that ask for almost nothing.